How we did it
We broke testing into several categories - Layer 2 switching, Layer 3 IP routing, IP Multicast, quality of service (QoS), virtual LANs, and STP support. We used Smartbits 2000 and Smartbits 6000 cards from Netcom systems for all testing. Visit www.netscomsystems.com for more information.
We used ML-7710 Fast Ethernet Layer 3 cards in the Smartbits 2000 chassis and LAN-6201A cards in the Smartbits 6000. For all tests, each Fast Ethernet port was set for full duplex, 100M bit/sec operation, and the Gigabit Ethernet ports were set for full-duplex operation. We disabled Ethernet autonegotiation for all ports on the switch under test.
For Layer 2 testing, we measured the throughput and latency of the switch with the transmitting and receiving ports in a pairwise bidirectional configuration, and the transmitter and receiver in a full mesh configuration. For the switches that have gigabit ports and Fast Ethernet ports, we did not run full mesh tests because the gigabit ports can easily overrun the Fast Ethernet ports. For Layer 3 testing, we measured the throughput and latency of the switch with the transmitting and receiving ports in a bidirectional pairwise manner. The pairwise throughput tests were run to measure the maximum throughput before loss. The latency values were calculated by measuring the time it takes to transmit the first bit of a packet from the transmitting Smartbits card and then receive the first bit back into receiving Smartbits card.
We conducted three tests to evaluate IP Multicast performance - mixed class throughput, forwarding latency, and group join/leave latency. Since these tests require multicast routing ability, we only ran these tests against the Catalyst 2948G-L3G-L3 and 4908F-L3G-L3. The mixed class throughput test looked at the packet loss of the device under test with equal amounts of unicast and multicast traffic within a multicast group. The unicast traffic was sent in a full mesh manner between receivers in a multicast group. This test analyzed how the switch reacted to multicast and unicast traffic occurring at the same time.
The multicast forward latency test measured the time it took to forward multicast traffic. The group join/leave latency test measured how long it took for the switch to start forwarding multicast traffic once an IGMP Join message was received from a group member, and how long it took the switch to stop forwarding multicast traffic once the last group member on a port sent an IGMP Leave message.
To test QoS, we ran two tests, depending on the device. For the Catalyst 3548XLXL, we ran a VLAN priority test. For the 2948G-L3G-L3 and 4908F-L3G-L3 we ran an IP precedence QoS test.
The VLAN priority test evaluated the switch's ability to prioritize traffic based on the 802.1Q priority bits in an Ethernet packet header. Once we found that the 3548XLXL did not properly prioritize traffic based on the VLAN priority bits, we modified the test to create a default priority per port.
The IP precedence test sent eight streams with equal numbers of packets for each possible IP precedence value. These streams were transmitted into two ports. Each of the streams were forwarded to a third port. The traffic was sent in varying amounts until the third port was overloaded. We measured the packet loss on the third port for each of the streams to evaluate if the streams were prioritized.
We tested VLANs on the 3548XLXL by evaluating VLAN functionality. We looked at VLAN by port, broadcast frames and illegal frame filtering.
The VLAN by port test ensured that VLAN traffic remained within the ports that were on that VLAN. The broadcast frames test verified that broadcast frames were contained within the ports that were members of a VLAN. The illegal frame filtering test evaluated the switch's ability to properly drop packets with bad cyclic redundancy check values, alignment errors, and packets that were too small or too big.
We measured the 3548XLXL's ability to reduce the STP convergence time using proprietary Cisco Spanning Tree optimization features. We tested the portfast feature by setting up all three switches in a triangle topology and measuring the time it takes for the network to recover when one of the active links between the switches goes down. The network switches to one of the redundant links. One this value is found, the network is reconfigured to enable portfast on the ports that aren't on links between the switches. The test is rerun to see that the convergence time is reduced.
RELATED LINKS
